this paper and patch for a multi-threaded cipher suite for ssh/scp on multi-core servers is very effective: http://www.psc.edu/networking/projects/hpn-ssh/
in regards to the proposed udp / dtls proposal, many of the same tricks apply to udp (socket buffer tuning, thread pools and event based socket handling (SEDA), etc.) is a new revision of the dtls proposal required before the datagram transport work is started? particularly reliable delivery via airhook or equivalent for existing TCP stream support? best regards,
